
KPI Definition Sheet: Prepare a Power BI Dashboard Brief
A KPI definition sheet is a short agreement describing what a performance indicator measures, how it is calculated, which records it includes and who maintains it. It gives the business owner and analyst a common reference before they build or approve a Power BI report or dashboard.
A metric becomes a key performance indicator when the organisation connects it to an important objective and a management decision. This article proposes a practical definition sheet and acceptance checks; it is not an official Microsoft template or evidence of HPA client performance.
Why agree the definition before drawing the chart?
Two charts can share a title and still answer different questions. One may count request records while another counts unique customers. One may use the creation date and another the completion date. If those choices are hidden, a visual difference may be mistaken for a business change.
Power BI adds another consideration: Microsoft’s measures documentation explains that calculated measure results can change as users interact with reports. Record the intended filters and test them rather than assuming a displayed total always describes the same population.
A working KPI definition sheet
| Field | Question to answer | Evidence or acceptance check |
|---|---|---|
| Decision and owner | Who will use the indicator, and what decision should it inform? | Name the business approver and intended action |
| Formula and unit | Is this a count, currency amount, duration or ratio? | Write the calculation in plain language, including numerator and denominator |
| Grain and identifier | What does one source record represent? | State the unique key and how duplicate records are resolved |
| Eligible population | Which statuses, locations and record types are included? | List exclusions explicitly; retain the counts excluded |
| Time basis | Which date field, period boundaries and time zone apply? | Check records at the beginning and end of a period |
| Source and access | Which approved system and fields supply the calculation? | Record the source owner and permitted audience |
| Filters and totals | Which filters may change the result, and how should totals aggregate? | Test overall and subgroup results against source counts |
| Freshness | What data period is available and when was it last updated? | Distinguish source availability from successful model refresh |
| Comparison and target | What prior period or approved target is comparable? | Retain the target owner, definition version and change date |
| Exceptions | What happens with missing fields or a zero denominator? | Show a documented unavailable or exception state; do not silently substitute success |
Start with one indicator and use the sheet as a discussion aid. Keep uncertain fields marked as open questions. Do not invent a target merely because the visual needs a red or green threshold.
Hypothetical example: combine a ratio using its counts
Illustrative data only, not HPA or client results. Suppose a service team wants to monitor the share of eligible requests containing all mandatory fields. Count each unique request once, exclude test records under a written rule, use the request creation date for the reporting period, and fix a reporting cutoff. The example assumes two non-overlapping groups and complete evidence.
| Group | Complete requests | Eligible requests | Rate |
|---|---|---|---|
| A | 90 | 100 | 90% |
| B | 1 | 10 | 10% |
| Combined | 91 | 110 | 82.7% |
Combined rate = 91 ÷ 110 × 100 = 82.7%, rounded to one decimal place. The simple average of the two subgroup rates is 50%; it gives the small group the same weight as the large one and does not represent this overall request-level rate.
Put the count logic in the definition sheet and check how report filters affect both numerator and denominator. The sheet should also state when completeness is assessed. A request completed after the reporting cutoff must not silently change a historical comparison without a documented restatement policy.
Specify freshness without promising real-time data
Microsoft’s data-refresh guidance distinguishes storage modes and their refresh requirements. In Import mode, the semantic model contains copied source data and needs refresh to fetch source changes. A polished visual does not establish that the source or the model is current.
As a proposed review rule, display both the source coverage date and the last successful model refresh when relevant. Assign an owner to investigate stale or incomplete data. A successful refresh can still contain an incomplete source period; refreshing a report is not proof that the operational team has finished entering records.
Acceptance checks before sharing the report
- Recalculate a small, approved sample from source evidence, including an excluded, duplicate and missing-field case.
- Test the unfiltered result and a relevant subgroup. Confirm the numerator, denominator and total follow the written rule.
- Check a period boundary, an empty period and a zero-denominator case. Ask the owner to approve the displayed exception state.
- Verify that the audience is permitted to see the detail, not just the headline metric. Use anonymised samples for training.
- Record approval, open limitations and a review date. Version changes to the definition so historical comparisons remain understandable.
Questions teams should settle
Is the definition sheet a substitute for DAX or a data model?
No. It describes the intended business meaning. The analyst still has to implement, secure and validate the model and calculations. A correct formula cannot resolve an undefined business rule on its own.
Should every operational metric be called a KPI?
No. Reserve KPI status for measures linked to priority objectives and decisions. Supporting counts can remain diagnostic metrics that explain the headline result.
What if two departments disagree on the definition?
Keep both definitions visible until the accountable business owner resolves the difference. If both are useful, name them distinctly and explain their populations rather than publishing two conflicting numbers with the same label.
Turn the brief into a learning task
For wider context, read HPA’s business analytics overview and business process automation guide. The published Data Analysis Course outline includes Excel, Power BI, data preparation, models, measures and dashboards. Confirm the current scope and entry requirements with the training team.
For a team need, use HPA’s corporate training enquiry page to share participant roles, group size, current Excel or Power BI experience and the reporting task that needs improvement. A practical learning brief could ask participants to define one indicator, reconcile a sample and explain a limitation. Do not include confidential source files in an initial enquiry.
Primary references and scope
Microsoft Learn: Measures in Power BI Desktop; Microsoft Learn: Data refresh in Power BI. Reviewed 15 November 2026. The sheet, test cases and numerical example are original editorial suggestions, not a case study or a guarantee of business results.


